Catalog to Trino Enrichment

A dataset read from the platform catalog says whether it can be queried, and where. fetch on a table or view's mcp:dataset: reference returns the catalog record with a query_availability block, so the one catalog read an agent makes before writing SQL already names the table to query and the connection that reaches it.

What Gets Enriched

Read	Enrichment
fetch mcp:dataset:(trino,<connection>,<catalog>.<schema>.<table>)	query_availability: whether the table is queryable, its query table, its connection, and its estimated row count

API and GraphQL operations are datasets too, but no query engine reads them, so their records carry no query_availability.

Example

Tool call: fetch with ref: "mcp:dataset:(trino,production,hive.sales.orders)"

The record carries the dataset, its columns and their curation, its lineage within three hops either way, and:

{
  "query_availability": {
    "available": true,
    "query_table": "hive.sales.orders",
    "connection": "production",
    "estimated_rows": 1500000
  }
}

A table the query engine cannot reach comes back with available: false and the reason in error, which is what the reader needs to know before writing a query.

How It Works

sequenceDiagram
    participant A as Agent
    participant P as Platform
    participant C as Platform catalog
    participant T as Trino

    A->>P: fetch mcp:dataset:(trino,production,hive.sales.orders)
    P->>C: Read the dataset, its columns and lineage
    C-->>P: Catalog record
    P->>T: Is the table there, and how many rows?
    T-->>P: Availability
    P-->>A: Record + query_availability

The availability lookup is bounded by a short timeout; a lookup that fails or runs out of time leaves the block off rather than failing the read.

Table Resolution

A catalog dataset is keyed by the connection that sees it and the path that connection uses, so a Trino dataset's reference already names the table: mcp:dataset:(trino,production,hive.sales.orders) is hive.sales.orders on the production connection. No mapping between catalog names and Trino names is configured or needed.

Configuration

query:
  provider: trino
  instance: production    # The Trino connection availability is checked on

Row Count Estimates

estimated_rows comes from Trino's table statistics. A table with no statistics has none; run ANALYZE on it in Trino to collect them.

Troubleshooting

A table reads as not queryable when it should be:

1. Verify the Trino table exists with the expected name.
2. Check that the query provider names the right connection.
3. Ensure the connection's Trino user may see the table.

Next Steps

• Trino to Catalog - The reverse direction
• Platform Catalog - What the catalog holds and how it is curated